CN105210088B - Using the signal processing of partial discharge method and device of neural network - Google Patents

Using the signal processing of partial discharge method and device of neural network Download PDF

Info

Publication number
CN105210088B
CN105210088B CN201380076458.9A CN201380076458A CN105210088B CN 105210088 B CN105210088 B CN 105210088B CN 201380076458 A CN201380076458 A CN 201380076458A CN 105210088 B CN105210088 B CN 105210088B
Authority
CN
China
Prior art keywords
impulse waveform
module
waveform
neural network
value
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201380076458.9A
Other languages
Chinese (zh)
Other versions
CN105210088A (en
Inventor
A·迪斯特法诺
R·坎德拉
G·菲丝赛丽
G·C·吉阿考尼亚
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Prysmian SpA
Original Assignee
Prysmian SpA
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Prysmian SpA filed Critical Prysmian SpA
Publication of CN105210088A publication Critical patent/CN105210088A/en
Application granted granted Critical
Publication of CN105210088B publication Critical patent/CN105210088B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
    • G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
    • G01R31/1263—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials of solid or fluid materials, e.g. insulation films, bulk material; of semiconductors or LV electronic components or parts; of cable, line or wire insulation
    • G01R31/1272—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials of solid or fluid materials, e.g. insulation films, bulk material; of semiconductors or LV electronic components or parts; of cable, line or wire insulation of cable, line or wire insulation, e.g. using partial discharge measurements
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
    • G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
    • G01R31/1245—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials of line insulators or spacers, e.g. ceramic overhead line cap insulators; of insulators in HV bushings
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
    • G01R31/14—Circuits therefor, e.g. for generating test voltages, sensing circuits
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/10—Interfaces, programming languages or software development kits, e.g. for simulating neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Abstract

Signal processing of partial discharge method (1100), including：The first discrimination standard in (1000) following standard is set：Discharge signal acquisition, the classification of discharge signal noise filtering, discharge signal；The associated multiple impulse waveforms of Partial Discharge signal that (1002) are provided and are detected；(1001) at least first are defined with reference to impulse waveform (RF) according to the first standard；Based on described at least first neural network module (800 is executed with reference to impulse waveform；900) the first training (1005), to assume that the similarity indices of first value and second value, first value and second value respectively represent similitude/dissimilarity of input pulse waveform and at least first reference impulse waveform (RF) with generating being suitably selected for property；Compare (1006) the multiple impulse waveform at least first with reference to impulse waveform, to obtain first similarity index value by neural network module；Based on the first similarity index value obtained and it is based on the second discrimination standard, memory/refusal (1008；1009) each impulse waveform through comparing.

Description

Using the signal processing of partial discharge method and device of neural network
Technical field
The present invention relates to signal processing of partial discharge method and devices.Shelf depreciation processing especially is used for analysis electrical part Shelf depreciation in part and system, electric component and system are such as：Medium-pressure or high pressure cable, cable connector, overhead line insulator, Medium and high pressure distribution box utilizes the high pressure and extra-high-tension cable of GIS (gas-insulated switchgear).
Background technology
Term " shelf depreciation " is intended to instruction in the presence of various types defect, the dielectric (insulation) of electronic unit The undesirable recombination of the charge occurred in material, eventually leads to dielectric failure.Herein, pulse current is in the part of dielectric material Middle generation simultaneously makes Electromagnetic Wave Propagation pass through the power supply or earth cable of associate power system, and radiates and pass through various surrounding mediums (dielectric material, metal, air etc.).
When executing measurement of partial discharge, a large amount of pulse signals are acquired and processed.Modern instrument allows with very high Sample frequency digitized pulse signal so that entire impulse waveform can be acquired and processed.The behaviour executed in measurement process Work is the selection to specific pulse according to scheduled discrimination standard.As an example, possible discrimination standard is：Discharge signal is adopted Collection, the classification of discharge signal noise filtering, discharge signal.
Discharge signal acquisition is related to some waves that only (have the level higher than specified threshold) in the waveform that detects The selection of shape.Discharge signal noise filtering is related to the selection to practical partial discharge pulse and refusal noise.Discharge signal point Class is related to the concrete property strobe pulse according to pulse and most like is grouped into different classes.
Discharge signal acquisition can be based on the frequency filtering and level threshold realized by analog circuit.And discharge signal noise Filtering and classification are executed by being typically based on the selection method of waveshape feature abstraction.These algorithms are usually by from each pulse They and specific threshold value are simultaneously compared to work by the parameter (feature) of waveform extracting a small group, so attempt to estimate each arteries and veins Whether punching falls into specific class.The validity of these algorithms depends critically on selected specific features group.
Selection method using neural network is known.Document JP02-296162 describes outer for detaching and detecting The method of portion's noise and local discharge signal.According to this method, the shielded layer of cable is cut, so obtain two it is opposite Cut end.The pulse-shaped voltage waveform generated at a cut end is by neural network and the generation at another cut end Another pulse-shaped voltage is compared.Two pulse-shaped voltage waveforms are all entered at the input layer of neural network.This ratio Relatively allow to distinguish local discharge signal (for this signal, similar pulse is generated at two cut end) and noise signal.
Document JP02-296161 disclose for by make the different wave that neural network learning is formed by shelf depreciation come The method for detecting partial discharge position.The electric unit that shelf depreciation generates wherein is result based on study and newly-generated The wave of shelf depreciation detects.
Document JP08-338856 illustrates the method that whether there is for determining shelf depreciation.The method comprising the steps of：Religion Lead neural network；Shelf depreciation is detected from the specified point of power cable and noise signal is multiple；The signal detected is fed to Neural network and determining assessed value；Multiple assessed values are averaging and are compared with threshold value.
Invention content
The technical problem to be solved by the present invention is to provide can be based on discrimination standard (such as acquisition, noise filtering, pulse point Class) analysis local discharge signal is in different environments the available processing method on different electric devices the problem of.
Applicants have discovered that for being adapted to detect for different electric devices using the method for Processing with Neural Network input signal The harmful shelf depreciation in varying environment, wherein neural network utilize the reference pulse defined according to selected discrimination standard Waveform is trained to and is configured as generating the similarity indices for selectively assuming first value and second value, first value Input pulse waveform is represented with second value and with reference to similitude/dissimilarity of impulse waveform.
According in a first aspect, the present invention relates to signal processing of partial discharge method, this method includes：
The first discrimination standard in following is set：Acquisition, noise filtering, the classification of shelf depreciation；
The associated multiple impulse waveforms of Partial Discharge signal for providing and detecting；
First is defined with reference to impulse waveform according to the first standard；
The first training of neural network module is executed with reference to impulse waveform based on described first, it is selectively false to generate The similarity indices of fixed first value and second value, first value and second value respectively represent input pulse waveform and first With reference to similitude/dissimilarity of impulse waveform；
It is similar to obtain first by more the multiple impulse waveform of neural network module with first with reference to impulse waveform Property index result；
Based on the first similarity index result obtained and be based on the first discrimination standard, memory/refusal each through than Compared with impulse waveform.
Advantageously, which further includes：
Second discrimination standard different from the first standard is set；
According to the second standard selection second with reference to impulse waveform；
Execute the neural network module second is trained, and selectively assumes what first value was worth with second to generate Similarity indices, first value and second value respectively represent another input pulse waveform to second with reference to the similar of impulse waveform Property/dissimilarity；
By more the multiple impulse waveform of neural network module at least partly with second with reference to impulse waveform, to obtain Obtain second similarity index result；
Based on the second similarity index result obtained and be based on the second discrimination standard, memory/refusal each through than Compared with impulse waveform.
Advantageously, neural network module is configured as realizing that activation primitive included in a grouping, the grouping include：It is non- Linear function, scalariform function, Sigmoid functions, identity function, Tangtoid functions, jump function, sign function, segmented line Property function.
Preferably, neural network module includes：
Multiple input port, each input port are configured as receiving and the pulse associating in the multiple impulse waveform Numeral sample；
At least one neurode with associated multiple weights, and
Output port is configured to supply similarity indices and is connected to the neurode as output nerve node.
" neurode " is intended to refer to the processing element of neural network, artificial neuron.
Advantageously, the neural network module has the structure of multilayer perceptron and includes having multiple hiding neuromeres The hidden layer of point, the multiple hiding neurode have the input for being connected to multiple input port.
" multilayer perceptron " be intended to refer to include in digraph multilayer neurode feed forward-fuzzy control, wherein often One layer is fully connected to next layer.Particularly, if hidden layer is used, the neural network module includes having to be connected to The single hidden layer of the output of output nerve node.
Preferably, the first discrimination standard is following one：Discharge signal noise filtering, discharge signal classification.
Preferably, neural network module includes single hidden layer and is software module.
Alternatively, neural network module has the structure of perceptron and includes being connected between input and output port Single neurode." perceptron " be intended to refer to include single neurode neural network.Particularly, the perceptron is used to Realize the discharge signal acquisition standard.Advantageously, the neural network module is hardware module.
Preferably, when executing the second training, first and second discrimination standard is discharge signal noise filtering and puts Electric signal is classified.
Advantageously, executing the first training includes：The multiple weight is calculated by providing counter-example to neural network module Multiple weighted values, for these counter-examples, request represents second value of dissimilarity result.
Preferably, by neural network module be compared including：
Multiple sample packets are defined, each sample packet represents the impulse waveform of the multiple impulse waveform；
Each sample packet is supplied to multiple input port；
Similarity indices value is obtained for each sample packet.
Advantageously, before each sample packet is supplied to multiple input port, this method further includes：
Execute waveform normalization, the wherein each sample divided by waveform bare maximum of sample packet；
Execute peak value alignment so that each peak value sample of each sample packet is provided to the choosing in multiple input port Fixed input port.
Particularly, define first includes with reference to impulse waveform：According to the first standard synthesis first with reference to impulse waveform.
Alternatively, define first includes with reference to impulse waveform：Selection first is with reference to pulse from the multiple impulse waveform Waveform.
Advantageously, the processing further includes：
The associated Partial Discharge signal of shelf depreciation of detection and electric component；
The synchronous electromagnetic signal that detection is generated by the associated alternating current voltage of operation with electrical object；
The Partial Discharge signal is synchronized to the synchronous electromagnetic signal.
According to second aspect, the present invention relates to signal processing of partial discharge systems, including：
Detection device, the associated multiple impulse waveforms of Partial Discharge signal for being structured to detection and detecting；
Processing module is connected to detection device and includes：
Neural network module, including：
Input module, the input pulse waveform for receiving the multiple impulse waveform；
Comparison module is connected to input module and is structured to compare with reference to impulse waveform and input pulse waveform； And
Output port is connected to comparison module and is structured to provide selectively hypothesis input pulse waveform and ginseng The similarity indices (SI) of first value or second value according to similitude/dissimilarity of impulse waveform, first value Or second value respectively represents the property of the impulse waveform detected,
The wherein described processing module includes：
Setup module is structured to discrimination standard of the setting among following：Discharge signal acquisition, discharge signal are made an uproar Sound filtering, discharge signal classification；And
With reference to selecting module, it is connected to setup module and is structured as according to selected discrimination standard selection reference arteries and veins Rush waveform.
Description of the drawings
By the description below with reference to attached drawing to the preferred embodiment and its alternative arrangement that provide as an example, further Characteristics and advantages will become more apparent, wherein：
Fig. 1 shows electrical object, includes with acquisition and analytical equipment and the shelf depreciation of processing module being selected to examine Survey the example of the shelf depreciation acquisition system of device.
Fig. 2 shows putting for can be used by the shelf depreciation acquisition system including high-pass filtering module and amplifier The example of electric treatment module；
Fig. 3 shows the example for the synchronous processing module that can be used by the shelf depreciation acquisition system；
Fig. 4 schematically shows the embodiments of acquisition and analytical equipment；
Fig. 5 a) shows the example of the trend of impulse waveform to be processed, and Fig. 5, b) show can be by the office First example of neural network module that portion's electric discharge acquisition system uses, with multilayer perceptron structure；
Fig. 6 is shown can be by the architecture for the exemplary neural node that the neural network module uses；
Fig. 7 a) again illustrates the example and Fig. 7, b of the trend of waveform pulse to be processed) show by It is embodied as second example of the network module of perceptron；
Fig. 8 shows the reality of the flow chart of the example for the processing method that explanation can be used by the shelf depreciation acquisition system Apply example.
Specific implementation mode
Fig. 1 shows that electrical object 100 and shelf depreciation acquisition system 500, wherein shelf depreciation acquisition system 500 include Local discharge detection device 400, acquisition and analytical equipment 300 and the choosing especially outside local discharge detection device 400 Select processing module 700.Acquisition and analytical equipment 300 may include also include local discharge detection device 400 outer cover in (such as In Fig. 1) or can be provided in individual outer cover, such as together with selection processing module 700.As an example, it locally puts Electric detection means 400 is portable and includes one or more battery.
Electrical object 100 can be any kind of component that can generate shelf depreciation electromagnetic pulse, unit or System, and be as an example：Medium-pressure or high pressure cable, overhead line insulator, medium-pressure or high pressure distribution box, makes cable connector With the high pressure and extra-high-tension cable, electro-motor or generator or medium-pressure or high pressure transformation of GIS (gas-insulated switchgear) Device.
Shelf depreciation acquisition system 500 is can be used to detection, measure and/or handle and analyze by as electrical object The electronic device for the shelf depreciation that 100 electrical source generates.Particularly, shelf depreciation acquisition system 500 can be portable And it is included in shell not shown in the figure.
Shelf depreciation acquisition system 500 is preferably configured as being placed near electrical object 100, is corresponded to by electricity with detection The electric discharge electromagnetic signal S for the partial discharge pulse that gas object 100 emitsd.It is also observed, may interfere with corresponding to shelf depreciation arteries and veins The electromagnetic noise signal S of the detection of the electromagnetic signal of punchingnIt can be deposited in the region for wherein using shelf depreciation acquisition system 500 .
The discharge signal S to be detecteddIt can be the electricity with the frequency being included in the range of 0.1MHz to 100MHz Magnetic wave pulse.Noise signal SnUsually there is the frequency being included in the same range of 0.1MHz to 100MHz.
Local discharge detection device 400 includes being adapted to detect for discharging (for simplicity, hereinafter also referred to as " detection device ") Signal SdSensor 1, but sensor 1 can also receive undesired electromagnetic noise signal Sn.Sensor 1 can be sensing contact Device or non-contact sensor.Contact sensor is put into contact with or close to electrical object 100, and non-contact or wireless sensor is suitable In execution remote detection, that is, be physically contacted without electric wire or cable connection source and sensor device and not.As an example, Remote detection can execute with a distance from signal source 1cm to 10m.The example of contact sensor is：Rogowsky sensors With coupler transformer type Magnetic Sensor.The example of noncontacting proximity sensor is：Magnetic field proximity sensor, acoustic sensor and pressure Electric transducer.
According to described embodiment, sensor 1 is the antenna that may be mounted to that in support construction 2 (as an example).Make For another example, antenna 1 can be one of following antenna：Small patch antennas, loop aerial, dipole and ultra-wideband antenna. Preferably, antenna 1 is hollow ball that is spherical and including conductive material, and conductive material is such as metal or polymeric material Material.As an example, spherical antenna 1 shows the diameter being included between 3 and 30cm, it preferably includes between 5 and 20cm.Especially Ground, antenna 1 can be similar to the antenna described in patent application WO-A-2009-150627.
Antenna 1 is configured as receiving discharge signal SdWith undesired noise signal SnAnd it converts them to defeated first Go out available reception electric signal S on terminal 3in1(such as electric current).The local discharge detection device 400 of Fig. 1 further includes discharge treatment Module 600 and synchronous processing module 200.Discharge treatment module 600 shows to be connected to the first of antenna 1 by the first conductor wire 4 The first input end 5 of leading-out terminal 3, and be structured to execute and receive electric signal Sin1Bandpass filtering and/or amplification, So that providing output discharge signal S at second output terminal 6d-out。
Fig. 2 be related to include the discharge treatment module 600 of high-pass filtering module 7 example, wherein high-pass filtering module 7 has It is connected to the corresponding of first input end 5 to input and be structured to removal low-frequency noise, such as with less than 0.1MHz frequencies The signal of rate.As an example, high-pass filtering module 7 may include the first capacitor C1 being connected in series with first resistor device R1. The output of high-pass filtering module 7 is connected to the first amplifier 8, wherein the first amplifier 8, which has, is connected to second output terminal 6 Corresponding output terminals.As an example, another filter (such as band logical, band resistance or low-pass filter (not shown)) may be coupled to The output of high-pass filtering module 7, to obtain the whole bandpass response with desired characteristic.
First amplifier 8, which has, is used for supply voltage V1The first feedback for terminal 9 and being connected to the second of ground terminal GND Feedback is for terminal 10.As an example, the first amplifier 8 shows the bandwidth including at least first antenna 1 (such as range from 0.1MHz To the bandwidth of 100MHz) bandwidth.
Include being configured as receiving at the second input terminal 11 to the synchronous processing module 200 in detection device 400, filtering Wave and the first synchronous electric signal S of amplificationsyn1, and provide second at third leading-out terminal 12 and synchronize electric signal Ssyn2.First is same Walk electric signal Ssyn1Represent the trend of AC (alternating current) voltage for being provided to tested electrical object 100.In general, AC voltages have It is included in 1Hz to the frequency between about 1000Hz.First synchronizes electric signal Ssyn1Can for example by by antenna 1 to by passing through The supply electromagnetic signal S that the voltage of electrical object 100 generatessupThe wireless and non-contact detecting that executes and obtain.
Second input terminal 11 is connected to the first conductor wire 4 with to receiving electric signal Sin1(further include the first synchronization electric signal Ssyn1) received.In this example, antenna 1 be designed to as capacitive coupled sensors operate, with from supply electromagnetic signal SsupDetection first synchronizes electric signal Ssyn1.Moreover, antenna 1 is designed to provide the AC (alternating current) for electrical object 100 with feedback The suitable capacitive coupling of voltage, by way of example, show suitable coupled surface.
Alternatively, first electric signal S is synchronizedsyn1It can be examined by the synchronous sensor 13 that may be connected to the second input terminal 11 Survey, synchronous sensor 13 such as wireless and non-contact detecting another antenna or put into electrical object 100 or with The another type of sensor that another electric component to be supplied to the identical voltage operation of electrical object 100 is in contact.
Fig. 3 shows the synchronous processing module 200 for including amplifier module 14 (such as high-gain buffer amplifier), this is put Big device module 14 has the input of the second input terminal 11 of connection and is connected to the 4th leading-out terminal of low pass filter blocks 16 15.High-gain buffer amplifier 14 also has the third feedback for supply voltage V1 for terminal 17 and is connected to ground terminal GND The 4th feedback for terminal 18.
As an example, high-gain buffer amplifier 14 is voltage amplifier and has the gain for being more than 100.Moreover, high Gain buffer amplifier 14 shows the input-output impedance more than 1MOhm and can have the overall bandwidth less than 1kHz. In this example, low pass filter blocks 16 include the second resistor being connected between the 4th leading-out terminal 15 and node 19 R2, and the second capacitor C2 for being connected between node 19 and ground terminal GND.Node 19 is connected to third leading-out terminal 12。
Acquisition and analytical equipment 300 and/or selection processing module 700 are configured as according to following at least one discrimination standard Execute output discharge signal Sd-outDifferentiation process：Discharge signal acquisition, discharge signal noise filtering and discharge signal classification.It is special Not, each criterion can by with suitably select to be compared to realize with reference to impulse waveform, that is, show to surpass Cross the amplitude of the pulse of threshold level.Discharge signal gatherer process allows to show with collected based on (as an example) pulse Predetermined properties select and store the impulse waveform for meeting specific criteria.Particularly, this discrimination standard is put by output Electric signal Sd-outIt is realized with the comparison with reference to impulse waveform.Show the impulse wave with the specific degrees similitude with reference to waveform Shape is collected (generating trigger event), and otherwise they will be rejected.
Discharge signal noise filtering process allows from output discharge signal Sd-outIn or from output discharge signal Sd-out Identification corresponds to the impulse waveform of practical partial discharge phenomenon and refuses noise waveform in derived signal.According to example, this The processing of kind of discharge signal noise filtering can be to the additional of the noise filtering that is executed in discharge treatment module 600.Particularly, This discrimination standard is by being realized with the comparison with reference to impulse waveform, wherein it is existing to show shelf depreciation with reference to impulse waveform As typical and different from noise signal shape.
Discharge signal assorting process allow by from output discharge signal Sd-outOr from output discharge signal Sd-outDerived letter The impulse waveform selected in number is grouped as multiple and different classes corresponding to different partial discharge phenomenons.As an example, First partial Discharge pulse class refers to due to local discharge signal caused by the defects of interior dielectric of electrical object, and the second shelf depreciation Pulse class refers to the defect on surface, and another class refers to corona discharge.Particularly, this discrimination standard be by with reference to arteries and veins It rushes the comparison of waveform to realize, wherein showing the typical shape of predetermined class with reference to impulse waveform.
Above-mentioned each discrimination standard can be by being included in acquisition and analytical equipment 300 and/or at selection The neural network module in module 700 is managed to realize.Neural network module can be configured as being sentenced according to a single standard implementation Other process.In this case, in order to realize the differentiation process according to two or three in discrimination standard listed above, respectively Using two or three individual neural network modules.Alternatively, single Neural module can be provided, with selectively real Existing more than one discrimination standard.
Discharge signal gatherer process can by acquisition and analytical equipment 300 execute, and discharge signal noise filtering process and Discharge signal assorting process can be executed by selection processing module 700.
With reference to this first example, the acquisition of Fig. 1 and analytical equipment 300 are configured as receiving output discharge signal Sd-outAnd Processing step is executed, this output discharge signal S is represented to generated-outMultiple numeral sample DS.Particularly, it acquires and analyzes Equipment 300 is configured as indicating output discharge signal S using the multiple samples being included between 32 and 256d-outPulse Waveform.As an example, acquisition and analytical equipment 300 are structured to execute analog-to-digital conversion, selection, acquisition and synchronization process step Suddenly.
Fig. 4 schematically shows the examples of acquisition and analytical equipment 300, and wherein equipment 300 can including optional broadband Amplifier 71 is programmed, wherein amplifier 71 has the input and connection for second output terminal 6 for being connected to discharge treatment module 600 To the corresponding output of analog-digital converter 72 (ADC).Acquisition and analytical equipment 300 further include control module 73, are such as structured In order to control broadband programmable amplifier 71 and from analog-digital converter 72 receive data field programmable gate array (FPGA).Broadband Programmable amplifier 71 can be programmed to by the shifted signal S provided by control module 73offWith gain signal SgaImparting is put Electrical output signal Sd-outDeviant and gain amplifier value so generate enlarged output signal Saout。
Broadband programmable amplifier 71 allows (as an example) range to change from the Continual Gain Actuator of about -5dB to+40dB. Analog-digital converter 72, which is structured the clock signal CK synchronizations to be generated by control module 73 and generates, will be sent to control The change data DTA of module 73.As an example, analog-digital converter 72 can be with 8 (bit) resolution ratio, 250,000,000 samples of conversion per second This.This sample frequency allows to acquire electric discharge electric signal S with the temporal resolution of 4nsd-out.It has been observed that most of shelf depreciation Pulse is usually than 0.5 μ s long.
Particularly, control module 73 includes processing unit 74 (PU) (such as microprocessor), memory 75 (M) (such as RAM (random access memory)) and synchronous logic module 76 (SINL).More particularly, memory 75 can be cyclic buffer.Place Reason unit 74 is connected to the timing module 87 (TM) for providing clock signal.
Synchronous logic module 76 is configured as receiving the second synchronizing signal Ssyn2, entrained timing information is therefrom extracted, Such as period of AC voltages and phase, and transmit information to processing unit 74.
Input/output end port 77 allows with shifted signal SoffWith gain signal SgaForm will be generated by processing unit 74 Output order Comm be transmitted to broadband programmable amplifier 71.
Control module 73 also with igniter module 78 (TRLM) and with address generation module 79 (ADD-GEN), Middle address generation module 79 is configured as generating under the control of processing unit 74 writes new data and reading in memory 75 It is stored in address necessary to the data in memory 75.
Igniter module 78 is structured to execute discharge signal gatherer process and is configured as only differentiating mark to meeting Accurate amplification output signal SaoutSample trigger the amplification output signal S to leaving broadband programmable amplifier 71aoutSample Memory.
Control module 73 further includes host interface module 80 (INTF), to allow data from neurode be transmitted to by with It is set to the transceiver 81 (TR) for exchanging data-/ command with selection processing module 700 by wired or wireless connecting line BD, such as USB/ ethernet transceivers.
Control module 73 can also have the extraction module 83 (for example, coprocessor CO-P) for being connected to processing unit 74. Extraction module 83 is configured as executing pulse characteristics extraction from the data being stored in memory 79, and especially pulse characteristics are real-time Extraction.The example of possible pulse characteristics by coprocessor extraction is：Peak value and polarity, phase, energy, the duration and The rough estimate of Weibull parameters.
According to described first example, igniter module 78 includes first nerves network module 800；First nerves net The example of the architecture of network module 800 is shown in figure 5b.First nerves network module 800 is configured to supply representative reference The similarity indices SI of similitude between signal and input waveform signal.Similarity indices SI can selectively assume to represent First value (for example, value 1) of affirmative correlation result or second value (for example, value 0) for representing dissimilarity result.
The first nerves network module 800 of Fig. 5 b includes multiple input Port IP1-IPNAnd multiple input Port IP1-IPN At least one hidden layer 801, the output nerve node ON of interconnection1With output port OP1(for example, single output port).In Fig. 5 b There is the specific first nerves network module 800 shown the structure of multilayer perceptron, multilayer perceptron to have single hidden layer 801.Hidden layer 801 includes multiple hiding neurode HN1-HNH, these neurodes, which have, is connected to multiple input port IP1-IPNInput and each of have be connected to output nerve node ON1Output.
Fig. 5 a) shows the amplification output signal S received at analog-digital converter 72aoutIn included impulse wave Shape PWsTrend example.Analog-digital converter 72 is structured to by sample SW1-SWNRepresent impulse waveform PWs.Particularly, Input port IP1-IPNNumber N be equal to sample SW1-SWNNumber.Each sample SW1-SWNIt is indicated by digital value.
Fig. 6 shows the neurode HN that can correspond to hidden layer 8011-HNHOne of or correspond to output nerve section Point ON1Exemplary neural node EN architecture.The architecture of neurode EN shows multiple input X1-Xn, biasing it is defeated Enter B (for example, being fixed to 1), multiple weight W0-Wn, summing junction 802 (Σ), activation primitive module 803 (f (Σ)) and output 804.Neurode EN is configured such that each input XiValue be multiplied by weight WiAnd it sums in summing junction 802.Nerve Node, which exports, to be calculated as the function f of this summation.
The activation primitive module 803 used in neural network module 800 can be nonlinear, scalariform function, such as Sigmoid activation primitives.Other adoptable activation primitives are：Identity function, Tangtoid functions, jump function, symbol letter Number and piecewise linear function.Compared with discontinuous function (step, symbol etc.), continuous function (for example, Sigmoid, Tangentoid and piecewise linearity etc.) realize better result.
According to the first embodiment of first nerves network module 800, each of hidden layer 801 hides neurode HN1-HNH With output nerve node ON1Including biasing input B and use piecewise linearity activation primitive.As an example, 4 hiding neuromeres Point HN1-HN4It can be used and realize 256 input port IP1-IPN。
The sample for reference that first nerves network module 800 can will be used as input sample by selection is grouped and selects The target output result of similarity indices SI is trained to.During training step, input data is applied to input port IP1- IPN, in the output port OP of first network module 8001The current criteria at place by calculating (proceeding to output from input), and with The difference (i.e. mistake) of target output is evaluated.
Then, weight W is iteratively adjusted0-Wn, to minimize mistake.Once mistake is sufficiently low, first nerves network mould Block 800 can provide correct output for the input each received.If first nerves network module 800 has hidden layer 801, then being known as the algorithm of " backpropagation " can be used to execute training.Since this algorithm pass through exporting result and towards defeated Come in adjust weight W before entering layer0-Wn, and for large-scale neural network, implement quite heavy.If do not hidden Layer exists, then training algorithm is comparatively simple and quick (that is, Hebb is regular, Delta is regular or Else Rule).In both feelings Under condition, training can all be completed in software, once and weight W0-WnEnd value it is available, they are just in hardware module by under It carries.This avoids the hardware realizations of drill circuit.
(show) that first nerves network module 800 is implemented as perceptron in fig.7b according to second example.Perceptron Be there is the neural network of single neurode, therefore the first nerves network module 800 of Fig. 7 b be similar in Fig. 5, b) in institute The first embodiment shown, but it does not include any hidden layer：Input port IP1-IPNIt is directly connected to output nerve node ON1.Particularly, neural network module 800 is the perceptron using linear activation primitive or piecewise linearity activation primitive.As example Son, piecewise linearity activation primitive 803 are step activation primitive and weight W0-WnIt is 1 potential coefficient.It has been observed that with linear activation The perceptron of function is the special case of ADALINE networks (adaline).ADALINE networks are monolayer neural networks.
Igniter module 78 is preferably realized by the first nerves network module 800 of second example (such as institute in Fig. 7 b Show).Particularly, the first nerves network module 800 of igniter module 78 is realized within hardware and weight is within hardware by reality It is now fixed coefficient W0-Wn.The instruction of first example shown in training process ratio Fig. 6 b of second example shown in Fig. 7 b Practice process to be easier and need less memory capacity.For example, impulse waveform PWsEach sample SWiIt can be by 1 to 8 Integer representation.Particularly, Hebb rules or Delta rules can be used to execute training process.
This allows the very big reduction of memory and calculating demand to storing weight.It is and complete when for triggering purpose MLP is compared, and preference brings sizable advantage.Since ANN must within hardware be realized by logic circuit, to handle Common high sampling rate (100 to 200MHz) in these applications, therefore the availability for simplifying circuit is vital.
Referring now to selection and processing module 700 (Fig. 1 and 4), it includes following module/equipment：It is set with acquisition and analysis Standby 300 exchange transceiver 701 (TR), another processing unit 702 (PU), memory module 703 (M), the display of data-/ command With interface equipment 704 (DYS) (such as keyboard and/or touch screen).Moreover, selection and processing module 700 includes at least second god Through network module 900, which is specifically configured as realizing：Discharge signal noise filtering and/or electric discharge letter Number classification.
Selection and processing module 700 also allows to receive multiple numeral sample DS together with received from acquisition and analytical equipment 300 Timing information, and generate Phase-Resolved Analysis pattern, in Phase-Resolved Analysis pattern, any numeral sample DS all with supply electromagnetism Signal SsupSuitably synchronous phase is associated.As an example, processing module 700 allows to show this Phase-Resolved Analysis pattern, wherein The peak swing of each pulse is drawn relative to corresponding phase value and/or the sample of impulse waveform is shown relative to the time Show.
According to example, nervus opticus network module 900 can be with the first nerves network mould with reference to described in figure 6,7 and 8 Block 800 is same or like.Particularly, discharge signal noise filtering and discharge signal classification can be by two corresponding neural networks Module realizes that each module, which was similar to nervus opticus network module 900 or nervus opticus network module 900, both to be used Classify again for realizing discharge signal in realization discharge signal noise filtering.Preferably, nervus opticus network module 900 is software Module and particularly it is similar to first nerves network module 800 and include hidden layer 801 shown in Fig. 5 b.As example Son classifies for noise filtering and discharge signal, impulse waveform PWSEach sample SWiCan by 1 to 8 integer representation simultaneously And coefficient W0-WnThere can be fixed points to indicate that (for example, 16 expressions) or floating number indicate (for example, 32 expressions).
Fig. 8 shows explanation by first nerves network module 800 or by nervus opticus network module 900 more than realization The flow chart of the example of a period of time of the discrimination standard listed adoptable processing method 1100.Processing method 1100 includes that setting walks Rapid 1000, wherein the first discrimination standard in following standard is set：Discharge signal acquisition, discharge signal noise filtering, electric discharge Modulation recognition.Setting steps 1000 can be by being included in acquisition and analytical equipment 300 and/or in selection processing module 700 It is correspondingly arranged module (such as software module) execution.Processing method 1100 further includes reference selection step 1001 (REF-SEL), In at least one selected according to the specific discrimination standard to be realized with reference to pulsed RF.Particularly, selection is with reference to pulsed RF set. With reference to pulsed RF can by user using be included in acquisition and analytical equipment 300 and/or selection processing module 700 in corresponding join It is manually selected according to selecting module (such as software component).
This can be especially designed (synthesis) with reference to pulsed RF set, will such as refer to the operation of igniter module 78 into one Step description.Reference noise filtering differentiates or criteria for classification, this can be stored in selection processing module 700 with reference to pulsed RF set Memory 703 in Phase-Resolved Analysis pattern (that is, multiple impulse waveforms) in selected, such as by the searching step in Fig. 8 What 1002 (RF-RETR) were indicated.Each can correspond to show during acquired numeral sample DS is grouped with reference to pulsed RF One numeral sample of the impulse waveform of some specific objects, position wherein in specific object such as Phase-Resolved Analysis pattern, certain Specific range, timestamp or the other parameters of a pulse characteristics value.Particularly, can be located at Phase-Resolved Analysis mould with reference to pulsed RF Pulse in the specific region of formula, the specific region have the short span from the point being directed toward by user in Phase-Resolved Analysis pattern From.This method can generate satisfactory as a result, because the arteries and veins usually found out in the same area of Phase-Resolved Analysis pattern Punching is from identical physical phenomenon and therefore has similar shape.It must be noted that the reference pulse in similar pulse The selection of RF is not particularly critical, because similarity relationships are transferable.
Preferably, processing method 1000 includes waveform normalizing step 1003 (WAV-NORM) and peak value alignment procedures 1004 (PK-ALGN).In waveform normalizing step 1003, each impulse waveform PWSEach sample divided by waveform bare maximum. Peak value alignment procedures 1004 pass through along input port IP1-IPNWaveform sample is posteriorly or anteriorly shifted to execute so that each pulse Each peak value sample of waveform is provided to multiple input Port IP1-IPNIn select input port, to be directed at all places The position of the maximum value for the impulse waveform managed.Waveform normalizing step 1003 (WAV-NORM) and peak value alignment procedures 1004 It (PK-ALGN) can be by the correspondence computing module that is run under the control of processing unit 74 and/or another processing unit 702 (such as software component) is realized.
Waveform normalizing step 1003 and peak value alignment procedures 1004 allow to eliminate in waveform due to scaling and time migration Caused by two nonessential differences, it is therefore desirable to first nerves network module 800 and nervus opticus network module 900 are less Complicated architecture executes processing.
Processing method 1100 includes training step 1005 (TRN), wherein first nerves network module 800 or nervus opticus Network module 900 is trained to using the selected training input data with reference to pulsed RF set and counter-example waveform sets is corresponded to. Counter-example waveform can be the impulse waveform of special design (synthesis), or can be pair of acquired numeral sample DS groupings The number of impulse waveforms of the Ying Yu except the region for the Phase-Resolved Analysis pattern that user's selection is selected with reference to pulsed RF set Sample.Training step 1005 is by input port IP1-IPNUpper offer trains input data, in output port OP1Upper calculating As a result and mistake is calculated about the desired value of similarity indices SI to be performed.Then, weight W0-WnIt adjusts with being iterated, with Just mistake is minimized.In training step 1005, when being used as training input data with reference to one of pulsed RF, desired value is set It is set to 1, when one of counter-example is used as training input data, desired value is arranged to 0.Training step 1005 can be by training mould Block (such as in the software component for acquiring with being provided in analytical equipment 300 and/or in selection processing module 700) management.It has been observed that When first nerves network module 800 or nervus opticus network module 900 are realized by perceptron (Fig. 7 b) and use single reference When pulsed RF, training step 1005 can be by selecting each weight W0-WnEqual to corresponding to the used sample with reference to pulsed RF The value of this grouping executes.It should also be noted that in this case, the perceptron of Fig. 7 b is grasped by executing cross-correlation relatively Make.Training step 1005 is executed using the sample obtained from waveform normalizing step 1003 and peak value alignment procedures 1004.
In calculating step 1006 (COMP), first nerves network module 800 or nervus opticus network module 900 are supplying Multiple numeral samples corresponding to impulse waveform to be processed are fed in step 1007 (PLS-SUPPL).With reference to acquisition standard, Impulse waveform is included in amplification output signal SaoutIn.If executing noise filtering process, impulse waveform is by trigger mould Block 78 acquires and is sent to the numeral sample DS of processing module 700.When executing assorting process, it is supplied to input port IP1-IPNImpulse waveform correspond to the sample of impulse waveform selected in application noise filtering process.Calculate step 1006 It is executed using according to waveform normalizing step 1003 and 1004 processed sample of peak value alignment procedures.
Based on the value of the similarity indices SI obtained by first nerves network module 800 or nervus opticus network module 900, Impulse waveform can be rejected (refusal step 1008, REJ) or can be selected and remember (memory step 1009, MEM).But It is that the impulse waveform being rejected can be remembered, to execute further analysis.If processing method 1100 realizes electric discharge letter Number acquisition standard or noise filtering standard, the then impulse waveform selected in remembering step 1009 can be stored in memory In 703.
Realize processing method 1100 specific embodiment shelf depreciation acquisition system 500 operation example below into Row description.
(Fig. 1) in operation, antenna 1 pick up discharge signal Sd, noise signal SnWith supply electromagnetic signal SsupAnd it generates Receive electric signal Sin1.Receive electric signal Sin1Be discharged processing module 600 filtering, therefore reduce frequency range 0.1MHz extremely Ingredient and generation outside 100MHz will be fed to the output discharge signal S of acquisition and analytical equipment 300d-out.Receive telecommunications Number Sin1Or the signal picked up by synchronous sensor 13 is sent to the second input terminal 11, and electric signal is synchronized to form first Ssyn1, which enters synchronous processing module 200.Synchronous processing module 200 amplifies and filters the first synchronization electric signal Ssyn1, Therefore it generates and is supplied to acquisition electric signal S synchronous with the second of analytical equipment 300syn2。
With reference to the operation of acquisition and analytical equipment 300, control module 73 executes configuration step and acquisition step.It is walked in configuration In rapid, acquisition parameter, such as first network of the gain of broadband programmable amplifier 71 and trigger logic module 78 are established The training and initialization of module 800.The training of trigger logic module 78 be by consider first synthesis prototype pulse (that is, The special pulse shape for designing and characterizing all desired characteristics) as being performed with reference to pulse.The prototype arteries and veins of this synthesis Punching can then be chosen one or more of multiple discharge signals acquired self and be replaced with reference to pulsed RF.
In acquisition step, processing unit 74, trigger logic module 78 and the management of address generation module 79, which correspond to, puts Big output signal SaoutStorage of the data in memory 75.When the igniter module 78 of realization processing method 110 is by god Through network module 800 in output port OP1When detecting that instruction dedicated pulse shape meets the similarity indices SI of acquisition standard, Corresponding waveform pulse is collected (that is, being stored in memory 75) in the form of multiple numeral sample DS and stopped further The acquisition of data.It has been observed that into the amplification output signal S of trigger logic module 78aoutCan be continue sample flow without It is fixed sample set.In this case, as being indicated arrow DST, passed through in fig.7b by each sample of displacement defeated Inbound port IP1-IPN(just as in a shift register), amplification output signal SaoutIt is fed into first nerves network module 800： As effective impulse PWSWhen in correct position, similitude output port OP1The similarity indices with value 1 are generated, to send out Signal trigger.
Processing unit 74 collects the timing information from synchronous logic module 76 and timing module 87, and will be stored in storage Multiple numeral sample DS in device 75 are sent to selection processing module 700 together with corresponding timing information.
Selection processing module 700 receives multiple numeral sample DS and timing information, and by nervus opticus network module 900, execute the discharge signal noise filtering of multiple numerical data DS by realizing processing method 1100.In order to execute electric discharge letter Number noise filtering utilize the corresponding ginseng for representing reference partial discharge pulse waveform before nervus opticus network module 900 It is trained according to sample packet.Nervus opticus network module 900 generates the similarity indices SI of assumed value 1 or 0, to respectively Indicate that specific pulse is considered partial discharge pulse and is also considered noise pulse.Based on by nervus opticus network The discharge signal noise filtering that module 900 is realized selects the storage of processing module 700 or shows multiple filtered numeral samples, These numeral samples show the affirmative similarity indices of the result as discrimination standard.
Nervus opticus network module 900 or other neural network modules are used to through application processing method 1100 come real The now discrimination standard about discharge signal classification.In order to execute discharge signal classification, utilized before nervus opticus network module 900 Another another reference sample grouping with reference to partial discharge pulse's waveform that representative belongs to predetermined class is trained.Nervus opticus net Network module 900 generates the similarity indices of assumed value 1 or 0, to indicate respectively that specific pulse may be considered that belong to this pre- Determine class still to must be regarded as not being included in this class.Identical nervus opticus network module 900 can be used to grouping and belong to In another kind of or subclass pulse.
It is such as clear for foregoing description, using the neural network module for being trained to calculating similarity indices, locally put Electric signal processing method 1100 can be used for different differentiation marks with reference to pulse by what change used in training step Standard is (such as：Pulse collection, impulse noise filter and pulse classification).This flexibility of described method be using show with just It takes the uncomplicated structure of formula and all compatible difficulty in computation of non-portable detection device to realize, to allow in different rings Described shelf depreciation acquisition system is used in border.

Claims (19)

1. a kind of signal processing of partial discharge method (1100), including：
The first discrimination standard in (1000) following standard is set：Acquisition, noise filtering, the classification of local discharge signal；
The associated multiple impulse waveforms of Partial Discharge signal that (1002) are provided and are detected；
(1001) at least one first are defined with reference to impulse waveform (RF) according to the first standard；
Neural network module (800 is provided；900), the neural network module (800；900) have and be configured to supply similitude Single output port (the OP of index1)；
Sample for reference by offer described at least one first with reference to impulse waveform is grouped into the neural network module (800； 900) neural network module (800 is executed；900) the first training (1005), selectively assumes first to generate The similarity indices of value and second value, first value and second value respectively represent input pulse waveform with it is described Similitude/dissimilarity of at least one first reference impulse waveform (RF)；
It is the multiple to compare (1006) that first sample by providing the multiple impulse waveform is grouped into neural network module Impulse waveform, with reference to impulse waveform, is referred to described at least one first with obtaining first similarity at the single output port Mark result；
Based on the first similarity index result obtained and it is based on first discrimination standard, memory/refusal (1008； 1009) each impulse waveform through comparing.
2. processing method as described in claim 1, further includes：
In the acquisition of local discharge signal, noise filtering, classification, setting second discrimination standard different from the first standard；
At least one second is defined with reference to impulse waveform according to the second standard；
It is grouped with reference to the other sample for reference of impulse waveform based on described at least one second and to execute the neural network mould Second training of block, selectively assumes the similarity indices of first value and second value to generate, described the One value and second value respectively represent another input pulse waveform with described at least one second with reference to impulse waveform (RF) similitude/dissimilarity；
Compared to neural network module by providing at least part of second sample packet in the multiple impulse waveform (1006) the multiple impulse waveform at least partly with described at least one second with reference to impulse waveform, it is similar to obtain second Property index result；
Based on the second similarity index result obtained and it is based on the second discrimination standard, memory/refusal (1008；1009) every A impulse waveform through comparing.
3. processing method as described in claim 1, wherein neural network module are configured as realizing included in a grouping Activation primitive, the grouping include：Nonlinear function, scalariform function, Sigmoid functions, identity function, Tangtoid functions, rank Jump function, sign function, piecewise linear function.
4. processing method as described in claim 1, wherein neural network module (800；900) include：
Multiple input port (IP1-IPN), each input port is configured as receiving and be closed with the pulse in the multiple impulse waveform The numeral sample of connection；
With associated multiple weight (W0-Wn) at least one neurode (ON1, HN1-HNH), and
Wherein, it is configured to supply the single output port (OP of the similarity indices1) be connected to as output nerve section Point (ON1) neurode.
5. processing method as claimed in claim 4, wherein the neural network module have the structure of multilayer perceptron and Including with multiple hiding neurode (HN1-HNH) hidden layer (801), the multiple hiding neurode have is connected to The multiple input port (IP1-IPN) input.
6. processing method as claimed in claim 5, wherein the neural network module (800；900) include defeated with being connected to Go out neurode (OPN1) output single hidden layer (801).
7. processing method as claimed in claim 5, wherein the first discrimination standard is following one：Discharge signal noise filtering, Discharge signal is classified.
8. processing method as claimed in claim 7, including the neural network module of single hidden layer (801) (800；900) it is software module.
9. processing method as claimed in claim 4, wherein neural network module (800；900) with perceptron structure and Including being connected to input port (IP1-IPN) and the single output port (OP1) between single neurode (ON1)。
10. processing method as claimed in claim 9, wherein the perceptron is used to realize discharge signal acquisition standard.
11. processing method as claimed in claim 10, wherein the neural network module (800；900) it is hardware module.
12. processing method as claimed in claim 2, wherein first discrimination standard and the second discrimination standard are discharge signals Noise filtering and discharge signal classification.
13. processing method as claimed in claim 4, wherein the first training of execution includes：
By providing counter-example to neural network module, the multiple weight (W is calculated0-Wn) multiple weighted values, it is anti-for these Example, request represent second value of dissimilarity result.
14. processing method as claimed in claim 4, wherein：
The first sample grouping includes multiple sample packets, and each sample packet represents the pulse in the multiple impulse waveform Waveform；And wherein
Compare (1006) the multiple impulse waveform includes with reference to impulse waveform with described at least one first：
Each sample packet is supplied to the multiple input port (IP1-IPN)；
Similarity indices value is obtained for each sample packet.
15. processing method as claimed in claim 14, wherein each sample packet is supplied to the multiple input port (IP1-IPN) before, this method further includes：
Execute waveform normalization (1003), each sample wherein in sample packet divided by waveform bare maximum；
Execute peak value alignment (1004) so that each peak value sample in each sample packet is provided to the multiple input terminal Mouth (IP1-IPN) in selected input port.
16. processing method as described in claim 1, defined in (1001) at least first include with reference to impulse waveform (RF)：
According to the first standard synthesis at least first with reference to impulse waveform.
17. processing method as described in claim 1, defined in (1001) first include with reference to impulse waveform (RF)：
At least first reference impulse waveform is selected from the multiple impulse waveform.
18. processing method as described in claim 1, further includes：
The associated Partial Discharge signal (S of shelf depreciation of detection and electric component (100)d)；
Synchronous electromagnetic signal (the S that detection is generated by the associated alternating current voltage of operation with electrical objectup)；
The Partial Discharge signal is synchronized to the synchronous electromagnetic signal (Sup)。
19. a kind of signal processing of partial discharge system (500), including：
Detection device (1,2,3,600), the associated multiple arteries and veins of Partial Discharge signal for being structured to detection and detecting Rush waveform；
Processing module (300；700) it, is connected to detection device and includes：
Neural network module (800；900), including：
Input module (IP1-IPN), the input pulse waveform for receiving the multiple impulse waveform；
Comparison module (801；ON1), it is connected to input module (IP1-IPN) and be structured to compare with reference to impulse waveform (RF) and input pulse waveform；And
Single output port (OP1), it is connected to comparison module (801；ON1) and be structured to provide selectively assume it is defeated Enter first value of impulse waveform and similitude/dissimilarity with reference to impulse waveform (RF) or the similarity indices of second value (SI), first value or second value respectively represent the property of the impulse waveform detected,
The wherein processing module (300；700) include：
Setup module (1000) is structured to discrimination standard of the setting among following：Discharge signal acquisition, discharge signal Noise filtering, discharge signal classification；And
With reference to selecting module (1001), it is connected to setup module (1000) and is structured as according to selected discrimination standard choosing It selects with reference to impulse waveform (RF).
CN201380076458.9A 2013-05-10 2013-05-10 Using the signal processing of partial discharge method and device of neural network Active CN105210088B (en)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/EP2013/059715 WO2014180508A1 (en) 2013-05-10 2013-05-10 Partial discharge signal processing method and apparatus employing neural network

Publications (2)

Publication Number Publication Date
CN105210088A CN105210088A (en) 2015-12-30
CN105210088B true CN105210088B (en) 2018-07-13

Family

ID=48485139

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201380076458.9A Active CN105210088B (en) 2013-05-10 2013-05-10 Using the signal processing of partial discharge method and device of neural network

Country Status (10)

Country Link
US (1) US10408873B2 (en)
EP (1) EP2994856B1 (en)
CN (1) CN105210088B (en)
AR (1) AR096219A1 (en)
AU (1) AU2013389280B2 (en)
BR (1) BR112015028010B1 (en)
CA (1) CA2909903C (en)
DK (1) DK2994856T3 (en)
ES (1) ES2760617T3 (en)
WO (1) WO2014180508A1 (en)

Families Citing this family (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
BR112016026330B1 (en) * 2014-05-16 2022-01-04 Prysmian S.P.A. PARTIAL DISCHARGE DETECTION DEVICE, AND METHOD TO ACQUIRE PARTIAL DISCHARGE SIGNALS
KR101787901B1 (en) * 2016-06-14 2017-11-15 엘에스산전 주식회사 Power equipment diagnostic apparatus
US11055615B2 (en) * 2016-12-07 2021-07-06 Arilou Information Security Technologies Ltd. System and method for using signal waveform analysis for detecting a change in a wired network
WO2018160924A1 (en) * 2017-03-02 2018-09-07 Rosemount Inc. Trending functions for partial discharge
CN107192929B (en) * 2017-06-21 2023-11-03 贵州电网有限责任公司电力科学研究院 GIS partial discharge signal source system for checking ultrahigh frequency monitoring system and use method
CN107942210A (en) * 2017-11-14 2018-04-20 国网上海市电力公司 The classification of transformer pulse electric current Partial Discharge and denoising method and system
JP7130542B2 (en) * 2018-12-12 2022-09-05 株式会社日立製作所 Partial discharge detection device and partial discharge detection method
JP7282000B2 (en) * 2019-08-30 2023-05-26 株式会社日立製作所 Partial discharge determination device and method
US11313895B2 (en) 2019-09-24 2022-04-26 Rosemount Inc. Antenna connectivity with shielded twisted pair cable
CN110768661B (en) * 2019-10-21 2022-10-28 中国科学技术大学 Phase-locked amplifier based on neural network
IT202000004303A1 (en) * 2020-03-02 2021-09-02 St Microelectronics Srl CIRCUIT FOR DETECTING PARTIAL DISCHARGES, CORRESPONDING DEVICE, SYSTEM AND PROCEDURE
CN111985528B (en) * 2020-07-06 2024-04-23 国网天津市电力公司 PDGAN-based cable partial discharge data enhancement method
CN112782538A (en) * 2020-12-25 2021-05-11 深圳供电局有限公司 Cable partial discharge detection method and device based on ultralow frequency power supply cut-off signal
CN112698170A (en) * 2021-01-18 2021-04-23 上海格鲁布科技有限公司 Ultrahigh frequency partial discharge detection anti-interference method based on time domain waveform
CN112949109B (en) * 2021-01-26 2023-04-07 西北核技术研究所 Method for predicting overhead line electromagnetic pulse conduction environment based on deep neural network
CN113447769A (en) * 2021-05-31 2021-09-28 华东电力试验研究院有限公司 PD pulse group manual rapid extraction system based on two-dimensional characteristic parameter plane
JP7412392B2 (en) * 2021-08-06 2024-01-12 株式会社日立製作所 Partial discharge determination device and method
CN113376494B (en) * 2021-08-16 2021-11-09 国网江苏省电力有限公司电力科学研究院 Method for detecting potential defects of fire hidden danger of cable
CN114167238A (en) * 2021-12-03 2022-03-11 广东电网有限责任公司广州供电局 Cable partial discharge detection method, device, system, equipment and storage medium
CN114065824B (en) * 2022-01-17 2022-04-19 陕西公众电气股份有限公司 Method for quickly identifying type of partial discharge signal
CN114777818B (en) * 2022-03-09 2024-03-15 西门子数控（南京）有限公司 Signal excitation unit of encoder and encoder
CN115166453B (en) * 2022-09-08 2023-01-24 国网智能电网研究院有限公司 Partial discharge continuous monitoring method and device based on edge real-time radio frequency pulse classification
CN116226647B (en) * 2023-05-06 2023-07-28 北京云道智造科技有限公司 Power cable partial discharge mode identification method, device, equipment and medium

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1567665A (en) * 1999-01-28 2005-01-19 株式会社日立制作所 Method of diagnosing partial discharge in gas-insulated apparatus and partial discharge diagnosing syatem for carrying out the same
CN1955749A (en) * 2005-10-26 2007-05-02 韩国电力公社 Broad band uhf local discharge signal and graphics generator
CN102221651A (en) * 2011-03-11 2011-10-19 太原理工大学 Fault on-line diagnosis and early warning method of flameproof dry-type transformer for mine
CN102809718A (en) * 2012-07-25 2012-12-05 华南理工大学 Ultra-high-frequency partial discharge signal identification method for gas insulated switchgear (GIS)

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2883352B2 (en) 1989-05-10 1999-04-19 一幸 合原 Partial discharge detection method
JP2877347B2 (en) 1989-05-10 1999-03-31 一幸 合原 Partial cable discharge detection method
JP3100757B2 (en) 1992-06-02 2000-10-23 三菱電機株式会社 Monitoring and diagnostic equipment
JP3107357B2 (en) 1995-06-12 2000-11-06 日立電線株式会社 Partial discharge determination method
FR2855878B1 (en) * 2003-06-05 2005-08-05 Alstom METHOD FOR DETECTING PARTIAL DISCHARGES AND DIAGNOSTIC SYSTEM FOR ELECTRICAL APPARATUS
ITRM20080304A1 (en) 2008-06-11 2009-12-12 Univ Palermo PORTABLE DEVICE FOR DETECTION OF PARTIAL DISCHARGES

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1567665A (en) * 1999-01-28 2005-01-19 株式会社日立制作所 Method of diagnosing partial discharge in gas-insulated apparatus and partial discharge diagnosing syatem for carrying out the same
CN1955749A (en) * 2005-10-26 2007-05-02 韩国电力公社 Broad band uhf local discharge signal and graphics generator
CN102221651A (en) * 2011-03-11 2011-10-19 太原理工大学 Fault on-line diagnosis and early warning method of flameproof dry-type transformer for mine
CN102809718A (en) * 2012-07-25 2012-12-05 华南理工大学 Ultra-high-frequency partial discharge signal identification method for gas insulated switchgear (GIS)

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
《Development of Partial Discharge Monitoring Technique Using a Neural Network in a Gas Insulated Substation》;I.Oki;《IEEE TRANSACTIONS ON POWER SYSTEMS 》;19770501;第12卷(第2期);第1014-1021页 *

Also Published As

Publication number Publication date
AU2013389280A1 (en) 2015-10-29
ES2760617T3 (en) 2020-05-14
AU2013389280B2 (en) 2019-11-21
CN105210088A (en) 2015-12-30
US20160116520A1 (en) 2016-04-28
WO2014180508A1 (en) 2014-11-13
EP2994856B1 (en) 2019-09-11
CA2909903A1 (en) 2014-11-13
US10408873B2 (en) 2019-09-10
CA2909903C (en) 2020-11-24
DK2994856T3 (en) 2019-12-09
EP2994856A1 (en) 2016-03-16
AR096219A1 (en) 2015-12-16
BR112015028010A2 (en) 2017-09-05
BR112015028010B1 (en) 2022-07-19

Similar Documents

Publication Publication Date Title
CN105210088B (en) Using the signal processing of partial discharge method and device of neural network
Hao et al. Partial discharge source discrimination using a support vector machine
CN106896306B (en) A kind of GIS oscillatory surge pressure test signal antinoise method
Gaing Wavelet-based neural network for power disturbance recognition and classification
Evagorou et al. Feature extraction of partial discharge signals using the wavelet packet transform and classification with a probabilistic neural network
CN108573225A (en) A kind of local discharge signal mode identification method and system
Peng et al. Rough set theory applied to pattern recognition of Partial Discharge in noise affected cable data
Chen et al. A novel extension neural network based partial discharge pattern recognition method for high-voltage power apparatus
CN109799434A (en) PD Pattern Recognition system and method
CN112716474A (en) Non-contact sleep state monitoring method and system based on biological microwave radar
CN107942210A (en) The classification of transformer pulse electric current Partial Discharge and denoising method and system
CN108693448B (en) Partial discharge mode recognition system applied to power equipment
WO2008143698A1 (en) Methods, systems and devices for detecting threatening objects and for classifying magnetic data
Zaro et al. Power quality detection and classification using s-transform and rule-based decision tree
Chang et al. Application of pulse sequence partial discharge based convolutional neural network in pattern recognition for underground cable joints
CN106503612A (en) Direct current cables local discharge signal mode identification method and system based on compressive sensing theory
Banno et al. Partial discharge source classification for switchgears with transient earth voltage sensor using convolutional neural network
CN115993511A (en) Partial discharge type high-precision detection and identification device, method and equipment
Swetapadma et al. A novel relaying scheme using long short term memory for bipolar high voltage direct current transmission lines
Hashim et al. Partial discharge localization in oil through acoustic emission technique utilizing fuzzy logic
CN105559771A (en) Electrocardiosignal quality detection method and apparatus
CN113449711A (en) Micro Doppler image sign language perception identification method based on direction density characteristics
Li et al. Detection of frequency-hopping signals based on deep neural networks
Aslan et al. ANN based fault location for medium voltage distribution lines with remote-end source
CN108153414B (en) Method, system and device for identifying steady-state visual evoked potential signals

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant